05-19-2021, 01:07 AM
Perhaps a form of conditioning or initial training might be possible, so that the AI has a bias toward obeying someone who has appropriate identification. Providing very positive rewards when performing desired action(s) during its initial development, perhaps. Since a “back door” might not be infallible, but might be more likely to result in a desirable reaction to a triggering event than if no such training had been used.
Finding appropriate rewards might be a challenge, of course.
Finding appropriate rewards might be a challenge, of course.
Selden